]> git.madduck.net Git - etc/vim.git/blobdiff - src/black/trans.py

madduck's git repository

Every one of the projects in this repository is available at the canonical URL git://git.madduck.net/madduck/pub/<projectpath> — see each project's metadata for the exact URL.

All patches and comments are welcome. Please squash your changes to logical commits before using git-format-patch and git-send-email to patches@git.madduck.net. If you'd read over the Git project's submission guidelines and adhered to them, I'd be especially grateful.

SSH access, as well as push access can be individually arranged.

If you use my repositories frequently, consider adding the following snippet to ~/.gitconfig and using the third clone URL listed for each project:

[url "git://git.madduck.net/madduck/"]
  insteadOf = madduck:

Better manage return annotation brackets (#2990)
[etc/vim.git] / src / black / trans.py
index 023dcd3618a5c0edb6b6f5e7e91e09bb6d3add30..01aa80eaaf84760f765ae34eef651d7336e28c09 100644 (file)
@@ -4,10 +4,11 @@ String transformers that can split and merge strings.
 from abc import ABC, abstractmethod
 from collections import defaultdict
 from dataclasses import dataclass
-import regex as re
+import re
 from typing import (
     Any,
     Callable,
+    ClassVar,
     Collection,
     Dict,
     Iterable,
@@ -20,6 +21,14 @@ from typing import (
     TypeVar,
     Union,
 )
+import sys
+
+if sys.version_info < (3, 8):
+    from typing_extensions import Literal, Final
+else:
+    from typing import Literal, Final
+
+from mypy_extensions import trait
 
 from black.rusty import Result, Ok, Err
 
@@ -62,7 +71,88 @@ def TErr(err_msg: str) -> Err[CannotTransform]:
     return Err(cant_transform)
 
 
-@dataclass  # type: ignore
+def hug_power_op(line: Line, features: Collection[Feature]) -> Iterator[Line]:
+    """A transformer which normalizes spacing around power operators."""
+
+    # Performance optimization to avoid unnecessary Leaf clones and other ops.
+    for leaf in line.leaves:
+        if leaf.type == token.DOUBLESTAR:
+            break
+    else:
+        raise CannotTransform("No doublestar token was found in the line.")
+
+    def is_simple_lookup(index: int, step: Literal[1, -1]) -> bool:
+        # Brackets and parentheses indicate calls, subscripts, etc. ...
+        # basically stuff that doesn't count as "simple". Only a NAME lookup
+        # or dotted lookup (eg. NAME.NAME) is OK.
+        if step == -1:
+            disallowed = {token.RPAR, token.RSQB}
+        else:
+            disallowed = {token.LPAR, token.LSQB}
+
+        while 0 <= index < len(line.leaves):
+            current = line.leaves[index]
+            if current.type in disallowed:
+                return False
+            if current.type not in {token.NAME, token.DOT} or current.value == "for":
+                # If the current token isn't disallowed, we'll assume this is simple as
+                # only the disallowed tokens are semantically attached to this lookup
+                # expression we're checking. Also, stop early if we hit the 'for' bit
+                # of a comprehension.
+                return True
+
+            index += step
+
+        return True
+
+    def is_simple_operand(index: int, kind: Literal["base", "exponent"]) -> bool:
+        # An operand is considered "simple" if's a NAME, a numeric CONSTANT, a simple
+        # lookup (see above), with or without a preceding unary operator.
+        start = line.leaves[index]
+        if start.type in {token.NAME, token.NUMBER}:
+            return is_simple_lookup(index, step=(1 if kind == "exponent" else -1))
+
+        if start.type in {token.PLUS, token.MINUS, token.TILDE}:
+            if line.leaves[index + 1].type in {token.NAME, token.NUMBER}:
+                # step is always one as bases with a preceding unary op will be checked
+                # for simplicity starting from the next token (so it'll hit the check
+                # above).
+                return is_simple_lookup(index + 1, step=1)
+
+        return False
+
+    leaves: List[Leaf] = []
+    should_hug = False
+    for idx, leaf in enumerate(line.leaves):
+        new_leaf = leaf.clone()
+        if should_hug:
+            new_leaf.prefix = ""
+            should_hug = False
+
+        should_hug = (
+            (0 < idx < len(line.leaves) - 1)
+            and leaf.type == token.DOUBLESTAR
+            and is_simple_operand(idx - 1, kind="base")
+            and line.leaves[idx - 1].value != "lambda"
+            and is_simple_operand(idx + 1, kind="exponent")
+        )
+        if should_hug:
+            new_leaf.prefix = ""
+
+        leaves.append(new_leaf)
+
+    yield Line(
+        mode=line.mode,
+        depth=line.depth,
+        leaves=leaves,
+        comments=line.comments,
+        bracket_tracker=line.bracket_tracker,
+        inside_brackets=line.inside_brackets,
+        should_split_rhs=line.should_split_rhs,
+        magic_trailing_comma=line.magic_trailing_comma,
+    )
+
+
 class StringTransformer(ABC):
     """
     An implementation of the Transformer protocol that relies on its
@@ -90,9 +180,13 @@ class StringTransformer(ABC):
         as much as possible.
     """
 
-    line_length: int
-    normalize_strings: bool
-    __name__ = "StringTransformer"
+    __name__: Final = "StringTransformer"
+
+    # Ideally this would be a dataclass, but unfortunately mypyc breaks when used with
+    # `abc.ABC`.
+    def __init__(self, line_length: int, normalize_strings: bool) -> None:
+        self.line_length = line_length
+        self.normalize_strings = normalize_strings
 
     @abstractmethod
     def do_match(self, line: Line) -> TMatchResult:
@@ -184,6 +278,7 @@ class CustomSplit:
     break_idx: int
 
 
+@trait
 class CustomSplitMapMixin:
     """
     This mixin class is used to map merged strings to a sequence of
@@ -191,8 +286,10 @@ class CustomSplitMapMixin:
     the resultant substrings go over the configured max line length.
     """
 
-    _Key = Tuple[StringID, str]
-    _CUSTOM_SPLIT_MAP: Dict[_Key, Tuple[CustomSplit, ...]] = defaultdict(tuple)
+    _Key: ClassVar = Tuple[StringID, str]
+    _CUSTOM_SPLIT_MAP: ClassVar[Dict[_Key, Tuple[CustomSplit, ...]]] = defaultdict(
+        tuple
+    )
 
     @staticmethod
     def _get_key(string: str) -> "CustomSplitMapMixin._Key":
@@ -243,7 +340,7 @@ class CustomSplitMapMixin:
         return key in self._CUSTOM_SPLIT_MAP
 
 
-class StringMerger(CustomSplitMapMixin, StringTransformer):
+class StringMerger(StringTransformer, CustomSplitMapMixin):
     """StringTransformer that merges strings together.
 
     Requirements:
@@ -268,7 +365,7 @@ class StringMerger(CustomSplitMapMixin, StringTransformer):
 
         is_valid_index = is_valid_index_factory(LL)
 
-        for (i, leaf) in enumerate(LL):
+        for i, leaf in enumerate(LL):
             if (
                 leaf.type == token.STRING
                 and is_valid_index(i + 1)
@@ -438,7 +535,7 @@ class StringMerger(CustomSplitMapMixin, StringTransformer):
             # with 'f'...
             if "f" in prefix and "f" not in next_prefix:
                 # Then we must escape any braces contained in this substring.
-                SS = re.subf(r"(\{|\})", "{1}{1}", SS)
+                SS = re.sub(r"(\{|\})", r"\1\1", SS)
 
             NSS = make_naked(SS, next_prefix)
 
@@ -473,7 +570,7 @@ class StringMerger(CustomSplitMapMixin, StringTransformer):
 
         # Build the final line ('new_line') that this method will later return.
         new_line = line.clone()
-        for (i, leaf) in enumerate(LL):
+        for i, leaf in enumerate(LL):
             if i == string_idx:
                 new_line.append(string_leaf)
 
@@ -594,7 +691,7 @@ class StringParenStripper(StringTransformer):
 
         is_valid_index = is_valid_index_factory(LL)
 
-        for (idx, leaf) in enumerate(LL):
+        for idx, leaf in enumerate(LL):
             # Should be a string...
             if leaf.type != token.STRING:
                 continue
@@ -739,7 +836,7 @@ class BaseStringSplitter(StringTransformer):
         * The target string is not a multiline (i.e. triple-quote) string.
     """
 
-    STRING_OPERATORS = [
+    STRING_OPERATORS: Final = [
         token.EQEQUAL,
         token.GREATER,
         token.GREATEREQUAL,
@@ -927,7 +1024,58 @@ class BaseStringSplitter(StringTransformer):
         return max_string_length
 
 
-class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
+def iter_fexpr_spans(s: str) -> Iterator[Tuple[int, int]]:
+    """
+    Yields spans corresponding to expressions in a given f-string.
+    Spans are half-open ranges (left inclusive, right exclusive).
+    Assumes the input string is a valid f-string, but will not crash if the input
+    string is invalid.
+    """
+    stack: List[int] = []  # our curly paren stack
+    i = 0
+    while i < len(s):
+        if s[i] == "{":
+            # if we're in a string part of the f-string, ignore escaped curly braces
+            if not stack and i + 1 < len(s) and s[i + 1] == "{":
+                i += 2
+                continue
+            stack.append(i)
+            i += 1
+            continue
+
+        if s[i] == "}":
+            if not stack:
+                i += 1
+                continue
+            j = stack.pop()
+            # we've made it back out of the expression! yield the span
+            if not stack:
+                yield (j, i + 1)
+            i += 1
+            continue
+
+        # if we're in an expression part of the f-string, fast forward through strings
+        # note that backslashes are not legal in the expression portion of f-strings
+        if stack:
+            delim = None
+            if s[i : i + 3] in ("'''", '"""'):
+                delim = s[i : i + 3]
+            elif s[i] in ("'", '"'):
+                delim = s[i]
+            if delim:
+                i += len(delim)
+                while i < len(s) and s[i : i + len(delim)] != delim:
+                    i += 1
+                i += len(delim)
+                continue
+        i += 1
+
+
+def fstring_contains_expr(s: str) -> bool:
+    return any(iter_fexpr_spans(s))
+
+
+class StringSplitter(BaseStringSplitter, CustomSplitMapMixin):
     """
     StringTransformer that splits "atom" strings (i.e. strings which exist on
     lines by themselves).
@@ -965,18 +1113,7 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
         CustomSplit objects and add them to the custom split map.
     """
 
-    MIN_SUBSTR_SIZE = 6
-    # Matches an "f-expression" (e.g. {var}) that might be found in an f-string.
-    RE_FEXPR = r"""
-    (?<!\{) (?:\{\{)* \{ (?!\{)
-        (?:
-            [^\{\}]
-            | \{\{
-            | \}\}
-            | (?R)
-        )+
-    \}
-    """
+    MIN_SUBSTR_SIZE: Final = 6
 
     def do_splitter_match(self, line: Line) -> TMatchResult:
         LL = line.leaves
@@ -1043,8 +1180,8 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
         # contain any f-expressions, but ONLY if the original f-string
         # contains at least one f-expression. Otherwise, we will alter the AST
         # of the program.
-        drop_pointless_f_prefix = ("f" in prefix) and re.search(
-            self.RE_FEXPR, LL[string_idx].value, re.VERBOSE
+        drop_pointless_f_prefix = ("f" in prefix) and fstring_contains_expr(
+            LL[string_idx].value
         )
 
         first_string_line = True
@@ -1284,9 +1421,7 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
         """
         if "f" not in get_string_prefix(string).lower():
             return
-
-        for match in re.finditer(self.RE_FEXPR, string, re.VERBOSE):
-            yield match.span()
+        yield from iter_fexpr_spans(string)
 
     def _get_illegal_split_indices(self, string: str) -> Set[Index]:
         illegal_indices: Set[Index] = set()
@@ -1402,7 +1537,7 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
         """
         assert_is_leaf_string(string)
 
-        if "f" in prefix and not re.search(self.RE_FEXPR, string, re.VERBOSE):
+        if "f" in prefix and not fstring_contains_expr(string):
             new_prefix = prefix.replace("f", "")
 
             temp = string[len(prefix) :]
@@ -1426,7 +1561,7 @@ class StringSplitter(CustomSplitMapMixin, BaseStringSplitter):
         return string_op_leaves
 
 
-class StringParenWrapper(CustomSplitMapMixin, BaseStringSplitter):
+class StringParenWrapper(BaseStringSplitter, CustomSplitMapMixin):
     """
     StringTransformer that splits non-"atom" strings (i.e. strings that do not
     exist on lines by themselves).
@@ -1578,7 +1713,7 @@ class StringParenWrapper(CustomSplitMapMixin, BaseStringSplitter):
         if parent_type(LL[0]) == syms.assert_stmt and LL[0].value == "assert":
             is_valid_index = is_valid_index_factory(LL)
 
-            for (i, leaf) in enumerate(LL):
+            for i, leaf in enumerate(LL):
                 # We MUST find a comma...
                 if leaf.type == token.COMMA:
                     idx = i + 2 if is_empty_par(LL[i + 1]) else i + 1
@@ -1616,7 +1751,7 @@ class StringParenWrapper(CustomSplitMapMixin, BaseStringSplitter):
         ):
             is_valid_index = is_valid_index_factory(LL)
 
-            for (i, leaf) in enumerate(LL):
+            for i, leaf in enumerate(LL):
                 # We MUST find either an '=' or '+=' symbol...
                 if leaf.type in [token.EQUAL, token.PLUSEQUAL]:
                     idx = i + 2 if is_empty_par(LL[i + 1]) else i + 1
@@ -1659,7 +1794,7 @@ class StringParenWrapper(CustomSplitMapMixin, BaseStringSplitter):
         if syms.dictsetmaker in [parent_type(LL[0]), parent_type(LL[0].parent)]:
             is_valid_index = is_valid_index_factory(LL)
 
-            for (i, leaf) in enumerate(LL):
+            for i, leaf in enumerate(LL):
                 # We MUST find a colon...
                 if leaf.type == token.COLON:
                     idx = i + 2 if is_empty_par(LL[i + 1]) else i + 1
@@ -1811,20 +1946,20 @@ class StringParser:
         ```
     """
 
-    DEFAULT_TOKEN = -1
+    DEFAULT_TOKEN: Final = 20210605
 
     # String Parser States
-    START = 1
-    DOT = 2
-    NAME = 3
-    PERCENT = 4
-    SINGLE_FMT_ARG = 5
-    LPAR = 6
-    RPAR = 7
-    DONE = 8
+    START: Final = 1
+    DOT: Final = 2
+    NAME: Final = 3
+    PERCENT: Final = 4
+    SINGLE_FMT_ARG: Final = 5
+    LPAR: Final = 6
+    RPAR: Final = 7
+    DONE: Final = 8
 
     # Lookup Table for Next State
-    _goto: Dict[Tuple[ParserState, NodeType], ParserState] = {
+    _goto: Final[Dict[Tuple[ParserState, NodeType], ParserState]] = {
         # A string trailer may start with '.' OR '%'.
         (START, token.DOT): DOT,
         (START, token.PERCENT): PERCENT,